AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Learning Rewrite-Invariant Reasoning with Targeted Alternation Training

Mousa Arraf, Ido Guy, Kira Radinsky

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add targeted alternation in-context demonstrations (200)GPT-5.2 (xhigh) · HLE (text only)solve rate (%): 35.4 -> 38.1Improved
generate meaning-preserving rewrites when building reasoning graphs (ablation)targeted alternation training · Big-Mathaccuracy (%): 66.7 -> 74.8Improved
generate meaning-preserving rewrites when building reasoning graphs (ablation)targeted alternation training · DROPaccuracy (%): 86.1 -> 91.9Improved
generate meaning-preserving rewrites when building reasoning graphs (ablation)targeted alternation training · MMLU-Proaccuracy (%): 76.6 -> 78.0Improved
use contrastive success/failure pairs instead of positive-only trajectories (ablation)targeted alternation training · DROP (GPT-4.1-mini)accuracy (%): 90.1 -> 91.9Improved
answer-only supervised fine-tuningDeepSeek-V2-Lite · MMLU-Proaccuracy (%): 60.1 -> 53.8Hurt
answer-only supervised fine-tuningGPT-4.1-mini, Phi-4, Llama-3.3-70B · MMLU-Proaccuracy (%): 76.5 -> 76.7 (GPT-4.1-mini)No clear effect
inference-time rewrite ensembling (majority vote over rewrites)Phi-4 · DROPaccuracy (%): 76.8 -> 82.2Improved
inference-time rewrite ensemblingGPT-4.1-mini · DROPaccuracy (%): 85.9 -> 85.0Hurt
hard-example mining via logit-margin uncertainty on rewritesLlama-3.3-70B-Instruct · MMLU-Proaccuracy (%): 71.9 -> 69.7Hurt
use GNN instead of logistic regression for the boundary-crossing predictor (ablation)boundary-crossing predictor · AUC across source modelsAUC: 70.43 -> 80.17 (DeepSeek)Improved
use GNN instead of graph transformer for the predictor (ablation)boundary-crossing predictor · AUC across source modelsAUC: 77.20 -> 83.83 (Phi-4)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy